Running an MCP Server

Running an MCP Server

An MCP server exposes tools and resources over a standard transport. You can run one locally for development or deploy it remotely for production agents.

Local Server

Run agent mcp serve --port 3001 to start a local MCP server. Your agent config can point to it via the mcp.servers array.

Remote Server

Deploy your MCP server to any HTTPS endpoint. Configure it in your agent with a URL and an optional API key for authentication.

Server transports

MCP servers communicate over one of two transports: stdio (standard input/output, for local processes) or HTTP with Server-Sent Events (for remote services). Queue supports both.

Running a local stdio server

For a local server, specify the command Queue should run to start the process:

mcp: servers: - name: my-local-tools command: node /path/to/my-mcp-server.js env: DATABASE_URL: ${DATABASE_URL}

Queue starts the process before the agent run begins and shuts it down when the run ends. The server communicates over stdin/stdout.

Running a remote HTTP server

For a remote server, specify the URL and authentication:

mcp: servers: - name: my-remote-tools url: https://mcp.my-service.com auth: type: bearer token: ${MY_SERVICE_TOKEN}

Developing a server

Use the official MCP TypeScript or Python SDK to build your server. The SDK handles protocol details so you can focus on defining tools and resources.

npm install @modelcontextprotocol/sdk

Testing your server

queue mcp inspect --server my-local-tools

This connects to your server and lists all available tools and resources, along with their schemas. Use it to verify your server is working correctly before running agents against it.